基于归并优化的快速特征点匹配算法

A Fast Method for Feature Point Matching Based on Merging Optimization

在使用点特征进行图像匹配的过程中,初始提取的特征点的数量和质量决定了最终的匹配速度和匹配精度。针对这一特点,本文提出对初始提取的特征点集首先进行动态的分块归并优化,然后使用改进的分层投影匹配算法对图像进行粗匹配,最后通过极线约束得到鲁棒的匹配结果。实验表明,本方法在对纹理丰富、尺寸较大的图像进行匹配时,大大提高了运算的速度,同时又保持了匹配的精度。

During the course of image matching by feature points, the number and the quality of the feature points which we picked up at the beginning work on the matching precision and rapidity. According to this trait, a merging method is proposed which is used to optimize the preliminary feature points, and then it takes advantage of hierarchy projection matching to filter feature points and chooses many possible matching points for one reference feature point. At last, it adopts constraint of epipolar to revise matching and removing false matching. Experiments prove this method is fast and reliable, especially for image matching with rich - texture and big size.